Abstract:

Aiming at the problems of the storage and transmission security of biometric templates,a cancelable tem-plate design method based on both random transformation and dynamic projection is proposed by taking into account the vector quantization (VQ)- based speaker recognition algorithm.In the enrollment stage,the LBG (Linde- Buzo-Gray) algorithm is adopted to train the voiceprint characteristics after random transformation,thus obtaining the codebook.Moreover,the codewords in the codebook are dynamically mapped to different random spaces.In the authentication stage,the voiceprint is mapped to the random space of the codewords by means of the same random transformation as that in the enrollment stage,and the verification is then carried out.Theoretical analyses indicate that the random transformation can improve the cancelability of the voiceprint template,and the dynamic projectioncan increase the uncertainty of the transformation.Simulated results show that the authentication performance of the system remains unchanged before and after the transformation,and that the system EER (Equal Error Ratio) is low when different random matrixes are used to respectively map the codewords and the voiceprint.The cancelability of the proposed method is thus verified.

Key words: random transformation, dynamic projection, cancelable template, authentication